Is there hierarchical generalization in response-effect learning?

Abstract

Ideomotor theory is an influential approach to understand goal-directed behavior. In this framework, response-effect (R-E) learning is assumed as a prerequisite for voluntary action: Once associations between motor actions and their effects in the environment have been formed, the anticipation of these effects will automatically activate the associated motor pattern. R-E learning is typically investigated with (induction) experiments that comprise an acquisition phase, where R-E associations are presumably learned, and a subsequent test phase, where the previous effects serve as stimuli for a response. While most studies used stimuli in the test phase that were identical to the effects in the acquisition phase, one study reported generalization from exemplars to their superordinate category (Hommel et al., Vis Cogn 10:965-986, 2003, Exp. 1). However, studies on so-called R-E compatibility did not report such generalization. We aimed to conceptually replicate Experiment 1 of Hommel et al. (Vis Cogn 10:965-986, 2003) with a free-choice test phase. While we did observe effects consistent with R-E learning when the effects in the acquisition phase were identical to the stimuli in the test phase, we did not observe evidence for generalization. We discuss this with regard to recent studies suggesting that individual response biases might rather reflect rapidly inferred propositional knowledge instead of learned R-E associations.

Keywords: Generalization; Ideomotor theory; Response-effect compatibility; Response-effect learning; Superordinate category.

Fig. 1

The upper panel illustrates the trial sequence and design of the acquisition phase, while the lower panel illustrates the trial sequence and design of the test phase. Stimulus words are presented here in the German language as in the experiment proper (translation: LOS! = GO!, MÖBEL = FURNITURE, TIERE = ANIMALS, KATZE = CAT, STUHL = CHAIR)

Fig. 2

Mean percentage of congruent choices in the category and the exemplar group. The dashed horizontal line at 50% indicates the expected value when response choice was entirely random and the error bars are the standard errors of the means

Fig. 3

Kernel density plots of the response biases separately for the category group and the exemplar group